Progressive Image Retrieval With Quality Guarantee Under MapReduce Framework

Because of the diversity and popularity of image acquisition techniques, data-driven methods for image analysis and editing have become popular. However, the explosive growth of images also presents challenges. Helping users to retrieve their expected images quickly and effectively is one of the most difficult tasks. Although various methods have been proposed, most methods cannot guarantee the quality of the image, which is typically required for analysis and authoring tasks. In this paper, we present a progressive image retrieval method with a quality guarantee. Images are gradually filtered by various criteria: starting from the quickest textual comparison, proceeding through a series of quality criteria, and ending with the most time-consuming contour match. The entire framework is parallelized under MapReduce to improve the performance. Various experiments are conducted to validate the performance and accuracy of the algorithm and the quality of the retrieved results. We also demonstrate the potential of the algorithm with an image synthesis prototype system.